[python] Preserve manifest order in Data Evolution splits

[XiaoHongbo-Hope]

Purpose

Native-plan CI exposed different Data Evolution split.files ordering between Python and Rust.

Java preserves the input manifest order within each overlapping row-id group (#6496). This PR aligns PyPaimon with that behavior by:

• removing the planner-only sequence pre-sort;
• preserving manifest order through slice/shard filtering.

The reader still applies its own merge ordering, so query results are unchanged.

Tests

• manifest order within one row-id group;
• mixed Blob/Parquet order after slice/shard filtering;
• external files with the same basename.

[JingsongLi]

+1